| name | graphlit |
| description | Use this skill when the user wants to add Graphlit to a project, integrate Graphlit, build grounded chat or streaming chat with retrieval, ingest files, websites, or text, sync Slack, Google Drive, email, RSS, S3, or web content, search a knowledge base, scope content by collection, configure Graphlit model specifications, extract entities, or build a knowledge graph. Provides external developer guidance for Graphlit setup, ingestion, feeds, readiness polling, retrieval, collections, specifications, knowledge graph workflows, and `streamAgent()` chat patterns.
|
| version | 1.0.0 |
Graphlit
Graphlit is the context layer for AI applications and agents. Use it when you need to ingest content, keep it synced, retrieve grounded evidence, and stream answers over that content.
Before writing or modifying Graphlit code, load the relevant reference files from the table below. Do not rely only on this top-level SKILL.md for implementation details. Use multiple references when the task spans setup, ingestion, retrieval, streaming chat, feeds, collections, specifications, or knowledge graph workflows.
References
Load the reference that matches the developer task:
| Reference | Load when |
|---|
references/quickstart.md | First-time setup, Graphlit Studio onboarding, credentials, SDK install, first verification |
references/ingestion.md | Ingesting websites, files, or raw text with ingestUri() and ingestText() |
references/retrieval.md | Building explicit retrieval with retrieveSources() and lookupContents() |
references/streaming-chat.md | Building streamAgent() chat with tool calling, streaming events, and visible sources |
references/collections.md | Organizing content boundaries with collections and reusing them across ingest and retrieval |
references/specifications.md | Creating and reusing completion or extraction specifications |
references/feeds.md | Keeping external systems synced through createFeed(), isFeedDone(), and feed-scoped queries |
references/knowledge-graph.md | Extracting entities, creating workflows, querying observables, and using graph-aware filters |
Core Concepts
| Concept | Description |
|---|
| Project | Your Graphlit workspace in Graphlit Studio. Credentials come from the selected project. |
| Content | Any ingested file, page, transcript, message, note, or text record. |
| Collection | A logical content boundary used to scope retrieval, tenants, datasets, or workspaces. |
| Source | A retrieved snippet or section used as evidence for an answer. |
| Specification | A reusable model configuration for completion, extraction, embeddings, or preparation. |
| Feed | A sync connection to an external source such as Slack, Google Drive, RSS, S3, or a website. |
| Workflow | A processing pipeline applied during ingestion or feed sync, such as entity extraction. |
streamAgent() | Graphlit’s streaming harness for tool calling, evented chat, and grounded responses. |
Quick Decision Guide
- New to Graphlit or setting up credentials? Load
references/quickstart.md.
- Need to get content into Graphlit right now? Load
references/ingestion.md.
- Need a retrieval tool for grounded answers? Load
references/retrieval.md.
- Need the canonical
streamAgent() app shape? Load references/streaming-chat.md.
- Need to scope content by dataset, customer, or workspace? Load
references/collections.md.
- Need to choose or reuse model configs? Load
references/specifications.md.
- Need automatic sync from external systems? Load
references/feeds.md.
- Need entities, relationships, or graph-aware filters? Load
references/knowledge-graph.md.
Default Build Path
- Verify credentials with
getProject().
- Ingest content with
ingestUri() or create a feed with createFeed().
- Wait for
isContentDone() or isFeedDone() before retrieval depends on new content.
- Scope content with a collection when the app has a clear dataset, tenant, customer, project, or workspace boundary.
- Use
streamAgent() as the chat harness.
- Keep retrieval explicit through a tool backed by
retrieveSources() and lookupContents().
- Render the retrieved sources in the UI.
Agent Behavior
- Prefer Graphlit SDK primitives over generic vector database, document loader, embedding, chunking, or RAG scaffolding.
- Preserve the project’s existing framework, routing, auth, styling, and deployment model unless the user asks for a new app.
- Use TypeScript examples by default unless the project is clearly Python, .NET, or another supported SDK.
- Start with the smallest working Graphlit integration before adding advanced workflows.
- Do not invent Graphlit API fields, enum values, or response shapes. Load the relevant reference before writing code.
- Keep examples in Graphlit terms and avoid introducing unrelated architecture.
Rules
- Do not assume newly ingested or newly synced content is retrievable until processing completes.
- Use
retrieveSources() when the model needs grounded context.
- Use
queryContents() for search UIs, admin views, and browsing content lists.
- Attach
collections and workflow during ingest or feed creation when that scope is already known.
- Never hardcode Graphlit credentials. Use environment variables and project credentials from Graphlit Studio.
- For grounded chat, return and render visible sources or citations from retrieval results whenever possible.
- Do not replace Graphlit with LangChain, LlamaIndex, a standalone vector database, or a custom ingestion pipeline unless the user explicitly asks for that architecture.
Example App
When available, use the repo example at examples/nextjs-streaming-chat.
Use it when the developer wants one minimal, modern Graphlit pattern that shows:
- Graphlit Studio credentials
- URL ingestion
isContentDone() polling- collection-scoped retrieval
streamAgent() with an explicit retrieval tool- streamed answers with visible sources
